Hi everyone, I'm Nicolai Tangen, the CEO of the Norwegian Sovereign Wealth Fund, and today I'm joined by Sridhar Ramaswamy, the CEO of Snowflake. Snowflake is basically the data platform that many of the world's biggest companies run on. So when your bank approves a loan or when a hospital pulls together patient data, Snowflake is often the engine underneath. Sridhar spent 15 years at Google where he built the advertising business from one and a half billion to over a hundred billion dollars. Then he walked away to start his own company and two years later he became the CEO of Snowflake. Now here at NBIM we are investors in Snowflake and we are also big users of the product. We have two petabytes of data in Snowflake which is the equivalent of 2 million gigabytes and we have roughly 3 million queries into the database every day. So big welcome, Sridhar.
Thank you, Nicolai. Happy to be here. Excited for the conversation.
First of all, how would you describe Snowflake to somebody who's never heard of it in brief?
Yeah, we are a data platform. We are like a cloud computing platform like an AWS, but with a strong focus on data. We help you do everything from bring in data from various different systems, analyze it, get insights from it, and then take it to the systems where you take action. So, we are an analytic data platform.
Gosh. Half the global 2000 companies that are addressable, that is non-China companies, are our customers. Hundreds and hundreds of customers in financial services, healthcare, advertising industries, the list goes on and on. And we operate out of more than 25 countries and have customers in way more than those.
When the company was founded in 2012, to separate storage from compute was pretty radical, right? Just tell us about it.
The simplest way to internalize that really important concept is to think about how you and I have always bought or used computers. I tell people for the past 50 years, whenever you and I wanted computing or our company, we would go buy a box and that box would have a fixed amount of storage, a fixed amount of compute, and a fixed amount of memory, and you're generally stuck with that box for I don't know 5 years. So if you decided that you wanted more memory, well too bad, you wait. If you decided that you bought too much compute, too much CPU power, well that's too bad. You already bought the machine. And so that's at the essence of it. By the way, your phone is another box, just a littler box. And so, and it was also really hard to move data from one box to another. That was an integration project. Snowflake radicalized things by sitting on top of cloud computing, which effectively are infinitely large databases with amazing amounts of compute that can be shared between different customers that operate there. And we separated out compute and storage. So we said you can use exactly the amount of storage that you want and the compute that you want. If you want to do like a deep analytic job because one of the people within NBIM came up with like this brilliant idea for investing that required you to go through every piece of historical information, you can spin up a thousand computers over one weekend, do the analysis, shut them down, and then figure out how to implement that in practice. That's the magic of Snowflake's model.
Most software companies charge per seat, but you charge for what people actually consume. Why? Why is that better?
Because it is aligned value creation. Part of the benefit that we have is we are very good at predicting demand because we have over 13,000 customers and so we are able to amortize spiky demand ourselves and offer you a more stable model that does not require you to make a commitment that you're going to use a certain amount of storage or a certain number of CPUs on an ongoing basis. And this in turn means that you spin up things as you need them. Most analytic jobs tend to be bursty and as you know the daylight variations even in transactional jobs is very high. People just are not doing as many banking transactions in the night and so there are all of these well-known patterns. What we are able to do with a consumption model is offer an incredibly effective price because we are able to amortize that and it also makes the company super value aligned. You pay for, we recognize revenue when you and your team use Snowflake and we'll come to this in the world of AI, that's a big deal because you only pay for what you use.
Absolutely. Sridhar, why do you now consider the likes of Anthropic and so on your competitors?
Because they're changing software as we know it. I think what is very, very clear to many of us working deeply in the software industry is that the cost economics of software are dramatically changing. And I would separate that out clearly from near-term concerns about costs of AI models and things like that. The reality is that software has always been a little bit of a craft for the past 50 years. Special people that are doing incredible amounts of work in order to produce software. Hard to build, hard to integrate. You should think of these like the best programmers are like concert pianists. There's no way to mint them. It's you spend the 10,000 hours and you have that special genius in your head and that's how you become an amazing musician. Software engineers have been like that. But what these models represent with software is the industrialization of software at a scale that we have not seen before. And that is the primary reason why I consider the model companies, things like the coding agents that they are creating, to be our biggest competition. This is the reason why we have a coding agent. They represent the front door to computing, the front door to all information more and more for everybody. Starts with software engineers, but it goes to more.
Absolutely. What about the likes of Amazon and Microsoft who also have competing products?
Yeah. You know, with all due respect to them, when it comes to pure data platforms, we have always been top of the charts when it comes to that. Companies that operate at that scale should never be underestimated because they are able to bring on just incredible resources to solve problems. But from what I see, I think of coding agents as the biggest threat to all software and figuring out a path for Snowflake that is going to survive and thrive in this world is my number one challenge.
Continuing on AI, in what ways are you benefiting from this revolution?
I take a reductive approach when it comes to AI and a software company like Snowflake. You know, at our core, we do two things. We create and run great software. This is what my engineering team does. We sell and help customers like you implement our software. So, a lot of our energy is focused very much on how do we make this go faster. On the sales side, our sellers now have tools that give them instant access to information. It's a sales agent that sits in their phone. Our solution engineers can do a custom demo for Nicolai in 30 minutes that has data that will look like it came from your bank. Their ability to use these tools, drive outcomes for you is incredible. And then on the software engineering side, coding is being revolutionized. I just came from a meeting where we are talking about how we need to get more people into spec-driven coding development. Which is you write an English language spec for what it is that you want to produce and then you automate the entire rest of the process of writing the first version of the code and testing it and deploying it and so on. And so software engineering is nothing like what it was two years ago. How to get every engineer in your team to act like that, to internalize these great possibilities is the biggest challenge that we have. But we have the superstars, people that are 50, 100 times more productive than the average software engineer.
How do large enterprises now change the data stacks on the back of AI?
So two things. First of all, AI makes the data that you have a lot more accessible. This is what products like Snowflake Intelligence do. It is roughly an agentic interface to your data where you can say how are my biggest investments doing? This investment seems to have gone down a little bit. Break it down for me. Is it because of the sector? Is it some other problem? All of those questions that required analysts to perform on your behalf are right now at your fingertips. And what this also means is that how work should be done in the future is also going to get dramatically changed because of this fast programmable access almost to the data. The second way in which a lot of our customers are benefiting is because of things like native coding agents. It's not often that I go to my customer and say I can make something that you have to do every day go 10 times or 20 times faster. That's what we are doing with internal teams and that's the value that we take to them. So both the builders and the consumers of data within a company have massive benefits thanks to AI.
Now we have something called an MCP which connects us to you guys. Could you just explain in short to the listeners what an MCP is?
Yeah, you should roughly think of that as, it stands for Model Context Protocol. It is just a way for a language model or a coding agent to be able to talk to a data source. And what we like our customers to do is use our products like Snowflake Intelligence that natively expose the data, but we're also all about being interoperable, working with the things that you folks have. So every agent that is created in Snowflake can be called remotely from some front door that you have and that's what MCP is. It is an interoperability layer.
How will agents change the way you work?
This is a big question and the word agents themselves is a little bit misunderstood. The way you should think about agents is it is a model and a piece of code that has access to different kinds of tools underneath and it knows how to call them smartly. So for example, if you ask it a question and say write a little program for me, it knows how to create a file, write some code into it and then call something locally to execute that piece of code. But exactly the same thing can be used if you say I want to know how my portfolio is doing. It knows how to call the portfolio tool, analyze it and give you back the result. And so agents are changing work at every level. Even though you tend to hear most about things like coding agents, but the general concept is very powerful. These coding agents are effectively abstraction agents. So even somebody that never wants to write a line of code is going to benefit enormously from having their basic functionality, access to all of your documents, access to the structured data that is in Snowflake. If I give you for example Snowflake Intelligence access to all of it, practically any question that you would want to ask of the data, the agent can come up with a plan and help you execute. That's why there's so much excitement about things like the work concept, cloud co-work, or Snowwork because they're changing how people think about work. Everything is programmable and also everything is interconnected. If you want to send an email based on an analysis you did, you don't have to cut and paste. You can just tell your agent, please send this email to Stefan and out goes the email.
Absolutely. Now, I'll tell you one thing. We spent years cleaning up our data and it's not a fun job. Now when you look at data, it's messy, duplicates, gaps, decades of legacy systems stitched together. Just what kind of barrier is that to the usage of AI and to the usage of your product?
Used to be a huge barrier to getting value from data. I think it is getting simpler by the day. That's the magic of coding agents applied to the data problem itself. Used to be that changing a pipeline for just adding one additional column of information in a complex data set that would easily be a week-long job for some poor programmer that had to toil through all of the details. We now have things like skills which you can think of as English language programs that automate that entire process from start to finish so that they can start it and they can come back in an hour and just ensure that everything has been taken care of. We are working on things like agent-driven migrations where you can move data from legacy systems onto Snowflake in a matter of days and small number of weeks as opposed to the multiple quarters and the multiple years that it used to take. And AI can be very, very powerful in the data modernization journey overall and it's an area that we are very heavily invested into. I wish we had it five years ago, but hey.
What about GDPR? How is that holding back usage, i.e. the personal data protection laws?
Well, part of Snowflake's benefit is that we offer excellent governance features. It is a burden for companies to make sure that they comply. You know, I'm American and everyone criticizes GDPR and it has had some unintended consequences. But there are also many aspects of GDPR that were genuinely forward-looking.
Is GDPR good or bad for Europe?
I think it's a very mixed bag. I think it had a lot of unintended consequences and it also tells you how regulation needs to be very surgical about what it does and people really need to think very hard. What are the unintended negative consequences? Yeah, I'll tell you the positive things first. The fact that you as a consumer can go to a company and say you need to delete everything that you know about me, that came as a result of GDPR. That is a huge positive. It forced every company. It forced my Google Ads team to make sure that we were actually able to track down everything that we knew about Nicolai. I think that's a huge positive. On the other hand, those walls of prompts that you have to, at this point you just give up and say yes. I think that's an unintended consequence. I think it has raised the cost of doing business for every European company and the people that ultimately benefited from GDPR were the giants who are able to spend the money, follow all the rules and be compliant while every new company that comes up in Europe has to struggle to follow all of these rules right out of the gate. That's what I mean by unintended consequences.
